Implementation of Inverse Reinforcement Learning Algorithm on a toy car in a 2D world problem, (Apprenticeship Learning via Inverse Reinforcement Learning Abbeel & Ng, 2004)
MIT License
176
stars
47
forks
source link
Weights for the reward function during the model learning #10
Hi @jangirrishabh,
This repository is awesome and well-explained. I want to thank you for the great content and code.
And I just have a question regarding the learning.py.
My question is: before training your NN model, you defined some weights in line:
https://github.com/jangirrishabh/toyCarIRL/blob/2eff036e594a787299d1e4cc82e46f0f9b21308f/learning.py#L206
and fed them into the carmunk to get the immediate reward and the new state based on the taken action to update the Y vector in the mini-batch process method. I was wondering how you defined the weights (weights for the reward function). Because later, you use this trained model in the toy_car_IRL.py to update the policy and reconstruct the weights for the reward function. So do those weights affect the trained NN model or they are just some random values?
Hi @jangirrishabh, This repository is awesome and well-explained. I want to thank you for the great content and code. And I just have a question regarding the learning.py. My question is: before training your NN model, you defined some weights in line: https://github.com/jangirrishabh/toyCarIRL/blob/2eff036e594a787299d1e4cc82e46f0f9b21308f/learning.py#L206 and fed them into the carmunk to get the immediate reward and the new state based on the taken action to update the Y vector in the mini-batch process method. I was wondering how you defined the weights (weights for the reward function). Because later, you use this trained model in the toy_car_IRL.py to update the policy and reconstruct the weights for the reward function. So do those weights affect the trained NN model or they are just some random values?